Koufaris and Nilsson Cancer & (2018) 6:12 https://doi.org/10.1186/s40170-018-0185-4

RESEARCH Open Access Protein interaction and functional data indicate MTHFD2 involvement in RNA processing and translation Costas Koufaris1,2,3 and Roland Nilsson1,2,3*

Abstract Background: The -coupled metabolic MTHFD2 is overexpressed in many tumor types and required for cancer cell proliferation, and is therefore of interest as a potential cancer therapeutic target. However, recent evidence suggests that MTHFD2 has a non-enzymatic function which may underlie the dependence of cancer cells on this protein. Understanding this non-enzymatic function is important for optimal targeting of MTHFD2 in cancer. Methods: To identify potential non-enzymatic functions of MTHFD2, we defined its interacting proteins using co-immunoprecipitation and mass spectrometry and integrated this information with large-scale co-expression analysis, protein dynamics, and expression response to MTHFD2 knockdown. Results: We found that MTHFD2 physically interacts with a set of nuclear proteins involved in RNA metabolism and translation, including components of the small ribosomal subunit and multiple members of the RNA-processing hnRNP family. Interacting proteins were also in general co-expressed with MTHFD2 in experiments that stimulate or repress proliferation, suggesting a close functional relationship. Also, unlike other folate one-carbon , the MTHFD2 protein has a short half-life and responds rapidly to serum. Finally, shRNA against MTHFD2 depletes several of its interactors and yields an overall transcriptional response similar to targeted inhibition of certain ribosomal subunits. Conclusions: Taken together, our findings suggest a novel function of MTHFD2 in RNA metabolism and translation. Keywords: Interactome, Moonlighting, Heat shock proteins, RNA, Non-metabolic

Background unclear, but may be due to the mitochondrial NADH/ Altered metabolism is a hallmark of cancer cells that NAD ratio favoring flux in the oxidative direction [8], or facilitates their growth and survival [1]. In particular, me- related to the associated reduction of NAD and NADP tabolism of folate-coupled one-carbon groups is repro- [7]. In the nucleus, the formate produced by the mito- grammed in a wide variety of cancers [2–4]. One-carbon chondrial pathway supports synthesis of dTMP during metabolism is localized into cytosolic, mitochondrial, and DNA replication. nuclear compartments (Fig. 1) and is required for the syn- Within mitochondrial folate one-carbon metabolism, thesis of , dTMP, and remethylation of homocyst- the enzyme MTHFD2 has attracted considerable attention eine to . Extensive evidence now supports that as a potential target for cancer therapeutics [4, 9, 10], mo- proliferating cancer cells primarily rely on the mitochon- tivated by a favorable expression profile with high expres- drial catabolism of serine via SHMT2 as their main source sion in various human tumor types [4]butlowor of one-carbon groups [5–7], rather than the correspond- undetectable levels in most adult tissues. The normal ing cytosolic enzymes. The reason for this preference is function of MTHFD2 may be in embryogenesis, since the enzyme has long been known to be highly expressed and * Correspondence: [email protected] essential during embryonic development [11], and a par- 1Cardiovascular Medicine Unit, Department of Medicine, Karolinska Institutet, alogous mitochondrial enzyme, MTHFD2L, is expressed SE-171 76 Stockholm, Sweden 2Division of Cardiovascular Medicine, Karolinska University Hospital, SE-171 76 in normal adult tissues [12]. Expression of MTHFD2 in Stockholm, Sweden tumors also correlates with poor disease outcome in Full list of author information is available at the end of the article

© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 2 of 14

nucleus cytosol MTHFD2? serine serine serine

SHMT1/2 SHMT1 SHMT2 glycine glycine glycine TYMS dtmp ch2-thf ch2-thf ch2-thf purines MTHFD2 MTHFD1 MTHFD1 nad(p)h cho-thf cho-thf cho-thf ALDH1L2 MTHFD1 MTHFD1 MTHFD1L nadph formate formate formate co2

Fig. 1 Schematic diagram of enzymes, metabolites, and compartmentalization of folate one-carbon metabolism. One-carbon metabolism enzymes are now known to be present and active in three distinct compartments, the nucleus, mitochondria, and cytosol, linked by the flow of formate between them. The function of MTHFD2 in the mitochondria is well understood, although its function in the nucleus is not known breast cancer [13], liver cancer [14], and acute myeloid co-expression pattern and the transcriptional responses leukemia (AML) [10]. Knockdown of MTHFD2 can in- to knockdown. Taken together, our results suggest a pre- hibit cancer cell proliferation in vitro [4, 10, 15] and re- viously unrecognized role for MTHFD2 in RNA metab- duces (but does not abolish) tumor growth in vivo [7, 10]. olism and translation. This combination of specific expression in transformed cells and severe cell phenotypes upon knockdown render Methods MTHFD2 a compelling potential drug target. Cell culture Nevertheless, several observations suggest that the HCT-116 cell lines were obtained from the National − − known metabolic function of MTHFD2 may not be the Cancer Institute. The MTHFD2 CRISPR / knockout sole explanation for its prominent role in cancer. First, (D2-KO) and parental wildtype (WT) HCT-116 cells supplementing cultures with formate, the primary product were obtained from the lab of Dr. Rabinowitz [7]. Cells of the mitochondrial pathway (Fig. 1), fails to rescue were grown in RPMI-1640 medium supplemented with MTHFD2 knockdown cancer cells [4, 10], indicating that 5% FBS (Life technologies) and Penicillin/Streptomycin another function of the MTHFD2 protein might be neces- (Gibco) and maintained at 37 °C and 5% CO2. sary for growth. Second, cells can rapidly adapt to loss of For serum starvation experiments, 200,000 cells were MTHFD2 by shifting to the cytosolic one-carbon pathway plated in six-well plates. The next day, the medium was [7], suggesting a degree of redundancy where flux through removed, the cells were washed with PBS and then the mitochondrial pathway is not strictly essential. Third, serum-free RPMI-1640 medium was added. Cyclohexa- knockdown of the SHMT2 enzyme, which should also mide was obtained from Sigma Aldrich (C4859). For the block the mitochondrial pathway, does not cause cell serum re-stimulation experiments, after 48 h of serum death unless glycine is removed [2, 16]. Moreover, the starvation, wells were washed with PBS and then fresh MTHFD2 protein was recently found to be present in the RPMI-1640 containing 20% FBS was added to the cells. nucleus at sites of newly synthesized DNA, and over-expression of catalytically inactive MTHFD2 drives Real-time PCR cell proliferation [17], suggestive of a role in signaling/ RNA was isolated from cells using the Qiagen RNeasy mini regulation rather than metabolism. Knockdown of kit and quantified using a Nanodrop ND-1000 spectropho- MTHFD2 also affects invasion and migration in vitro [14, tometer. The RNA was reverse transcribed to cDNA using 18], although no obvious mechanistic link exists between the VILO cDNA synthesis kit (Invitrogen). The qPCR one-carbon metabolism and these phenotypes. reactions were performed in triplicate on a StepOne Given these observations, it is becoming clear that the Real-Time PCR machine (Thermo Fisher Scientific) using effective targeting of the MTHFD2 protein requires a the Fast SYBR Green master mix (Thermo Fisher better understanding of its role in cancer cells beyond Scientific). Oligos were ordered from Sigma Aldrich. Primer metabolism. Although chemical inhibitors of the sequences are listed in Additional file 1. Relative amounts MTHFD enzymatic activity have been reported [19, 20], of mRNA were calculated using the ΔΔCT method normal- different strategies may be required if a non-enzymatic ized to RPLPO mRNA as a reference. function of the MTHFD2 enzyme is critical for cancer cells. For these reasons, we performed an initial investi- Immunoblotting gation of the possible non-metabolic functions of Total protein was isolated from cells using RIPA buffer MTHFD2 by mapping the protein’s interacting partners, (Thermo Fisher Scientific), with the addition of × 100 Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 3 of 14

Halt protease inhibitors (Thermo Fisher Scientific) just Q Exactive HF Orbitrap mass spectrometer, Thermo prior to the extraction. Protein from nuclear and cyto- Scientific). The peptides were separated on an Easy-C18 solic cellular compartments were fragmented as de- column, 50 cm (Thermo Fisher Scientific) at 55 °C in a scribed previously [17]. Protein levels were quantified by 120-min gradient at a flow rate of 300 nL/min. The lin- the BCA assay (Pierce). Equal amounts of protein (10– ear gradient was from 5 to 26% of buffer B (98% aceto- 20 μg) were loaded per lane of 10% SDS-PAGE gel. The nitrile/0.1% formic acid) in 115 min and to 95% of following primary antibodies were incubated overnight: buffer B in 5 min. anti-MTHFD2 (Proteintech 12270-1-AP); anti-MTHFD1 We used data-dependent acquisition with a survey scan (Proteintech 6113-1-AP); anti-MTHFD1L (Proteintech range of 300 to 1650 m/z, at a resolution of 120,000 m/z 10794-1-AP); Lamin A/C (Thermo Fisher Scien- and selected up to 16 most abundant features with a tific MA3-1000); HSP60 (Abcam ab46798); Tubulin charge state ≥2 for HCD fragmentation at a normalized (Abcam ab6046); and COX IV (Abcam 33985). Quantifi- collision energy of 26 and a resolution of 30,000 at m/z cation of protein bands was performed using the ImageJ 200. To limit repeated sequencing, dynamic exclusion of software and values were normalized to tubulin for each sequenced peptides was set to 90 s. Thresholds for ion in- sample. jection time and ion target values were set to 250 ms and 5 × 106 for the survey scans, and 120 ms and 2 × 105 for Co-immunoprecipitation the MS/MS scans. Data were acquired using the Xcalibur WT and D2-KO HCT-116 cells were lysed using the mod- software (Thermo Fisher Scientific). The spectra were ana- erate Pierce IP lysis buffer (Thermo Fisher Scientific) with lyzed using the Mascot search engine v. 2.4 (Matrix Sci- addition of × 100 Halt protease inhibitors (Thermo Fisher ence Ltd., UK) using carbamidomethylation (C) as fixed Scientific). All steps were performed at 4 °C to reduce pro- and deamidation (NQ), oxidation (M) as variable modifi- tein disassociation. The protein lysate was quantified using cation. The precursor mass tolerance was set to 10 ppm the BCA assay, and 1 or 2 mg of lysate was used for im- while MS/MS tolerance was 0.02 Da allowing 2 missed munoprecipitation with no cross-linkage. Immunoprecipi- cleavages. The SwissProt database was used with tax- tation was performed using two separate anti-MTHFD2 onomy limited to Homo sapiens. antibodies (Proteintech 12270-1-AP (AP) and Genetex In total, we performed 12 Co-IP experiments (for N1C3 (NC)) or an anti-rabbit immunoglobulin (IgG) HCT-116 WT, 4 using AP antibody, 2 using NC anti- (Santa Cruz). The immunoprecipitation was performed at body, and 2 using IgG antibody; for HCT-116 D2-KO, 2 4 °C for 2 h using Dynabeads Protein A (Thermo Fisher using AP antibody and 2 using NC antibody) followed Scientific) following the manufacturer’s instructions. by MS analysis. A protein was considered as high confi- dence if at least two unique peptides were detected in In-gel digestion and mass spectrometry 50% of IP-MS samples for each of the MTHFD2 anti- Immunoprecipitated proteins were eluted from Dyna- bodies used and was not detected in any of the IP-MS beads through a combination of elution buffer and heat- samples from D2-KO cells, nor the IgG isotype control ing and separated on a 10% SDS gel. In-gel staining of IP-MS samples. proteins was performed using the SilverQuest silver staining kit (Thermo Fisher Scientific). Protein lanes Protein-protein interaction network were excised manually and in-gel digested using a A protein-protein interaction network was extracted MassPREP robotic protein-handling system (Waters, from the STRING database v10.5 [21] using the set of Millford, MA, USA). Gel pieces were distained twice MTHFD2 interacting proteins identified here. Only ex- with 100 μL of 50 mM ammonium bicarbonate (AmBic) perimental data (e.g., from co-purifications and yeast containing 50% acetonitrile at 40 °C for 10 min. Proteins two hybrids) imported in STRING from primary sources were reduced with 10 mM DTT in 100 mM Ambic for were used for constructing the network. Threshold for 30 min at 40 °C and alkylated with 55 mM iodoaceta- connections between proteins was set at medium score mide in 100 mM AmBic at 40 °C for 20 min followed (more than 0.400) as calculated in STRING, and gene with in-gel digestion with 0.3 μg trypsin (sequence ontology (GO) enrichment was calculated. grade, Promega, Madison, WI, USA) in 50 mM at 40 °C AmBic for 5 h. The tryptic peptides were extracted with Half-life analysis 1% formic acid in 2% acetonitrile, followed by 50% For half-life analysis, we used publicly available data on acetonitrile twice. The liquid was evaporated to dryness, global protein half-life in NIH3T3 cells [22], HeLa, and and the peptides were reconstituted in 10 μLof2% C2C12 cells [23]. The list of 1652 proteins involved in acetonitrile/0.1% formic acid. Five microliters of the metabolism were obtained from Recon2 [24] model of samples was injected onto the LC-MS/MS system human metabolism, of which 596 were measured in (Ultimate™ 3000 RSLCnano chromatography system and HeLa, 575 in C2C12, and 605 in NIH3T3. Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 4 of 14

Co-expression analysis analysis based on the KEGG ribosome gene set (entry Calculation of gene co-expression was performed using hsa03010). a clustering-based method, as described previously [17]. Briefly, we analyzed 25,485 represented on a var- RNAi and CRISPR screening data analysis iety of microarray platforms for co-expression with Global RNAi and CRISPR screening data were per- MTHFD2 across 8097 human, mouse and rat data sets formed as part of the Achilles project in 501 and 342 cell obtained from the Gene Expression Omnibus (GEO). lines respectively [28, 29]. We also examined a second From the resulting matrix of coexpression scores xgd,we CRISPR study conducted in 14 AML cell lines [30]. To computed the overall coexpression score wg for each identify genes with similar dependencies across exam- gene g by simply summing over all data sets d. We se- ined cell lines in the RNAi and CRISPR screens, we cal- lected the 50 data sets showing highest coexpression for culated Pearson’s correlation for MTHFD2 against all the D2PPI genesX by scoring each data set by the examined genes in both datasets. weighted sum wg xgd , where g ranges over the 29 g Results genes in the D2PPI gene set; and similarly for the ATF4 Protein interaction partners of MTHFD2 gene set (Fig. 3b). Example data sets from GEO (Fig. 3c–f) Since a non-metabolic function of MTHFD2 would were analyzed as deposited on GEO, with no further likely involve direct physical contact with other proteins, normalization. In cases where multiple probes against a we decided to first perform co-immunoprecipitation single gene were present on the arrays, the probe with (Co-IP) of MTHFD2-interacting proteins followed by highest mean signal across all samples was used. Enrich- mass spectrometry (MS) to identify its binding partners. ment analysis was done using the GSEA-P statistic [25], To avoid potential artifacts due to unspecific antibody and an enrichment p value was calculated by gene permu- binding, we chose to perform Co-IP experiments in tation (10,000 permutations). An independent analysis wildtype (WT) and CRISPR MTHFD2 knockout was made using the SEEK tool (http://seek.princeton.edu) (D2-KO) HCT-116 cells [7], using two distinct MTHFD2 with default settings [26]. antibodies, 12270-1AP (AP; Proteintech) and N3C3 (NC; Genetex) as well as an isotype control (IgG) anti- body (Fig. 2a). We first confirmed by immunoblotting Connectivity Map (CMap) analysis and IP that the MTHFD2 protein was absent from Expression data from the Connectivity Map (CMap) pro- D2-KO cells (Additional file 2). Next, we performed ject in Level 5 (signature) GCTX format was obtained Co-IP followed by SDS-PAGE and silver-staining to ver- from GEO (accession GSE92742), and data for 99 experi- ify the efficient immunoprecipitation of MTHFD2. This ments with five distinct shRNA hairpins against MTHFD2 analysis revealed a strong band at the expected size of across 12,328 measured genes was extracted using R MTHFD2, as well as a large number of other v.3.3.3 and the cmapR package (https://github.com/cmap/ immunoprecipitated protein bands (Fig. 2b). In D2-KO cmapR). Since the three hairpins TRCN0000036550, cells, the MTHFD2 band was not detected, but other TRCN0000036551, and TRCN0000036553 exhibited protein bands were still observable, presumably due strongest knockdown on MTHFD2 itself, these were used to non-specific binding of the antibodies used for subsequent analysis. Data from the ASC and NPC cell (Additional file 2). Detected peptide fragments in all ex- lines were discarded due to few replicates. The z-scores amined Co-IP lysates are listed in Additional file 3. for the remaining 9 cell lines were averaged into a final Reassuringly, MTHFD2 itself was reliably identified from z-score vector, which was used for enrichment analysis WT cells using both antibodies and strongly decreased in using the GSEA-P statistic [25]. The genesets examined the D2-KO lysates (Fig. 2c). were downloaded from ConsensusPathDB collection We defined high-confidence MTHFD2-interacting pro- (http://cpdb.molgen.mpg.de/). An enrichment p value was teins as those that were reliably detected by Co-IP in WT calculated by gene permutation (10,000 permutations). but not D2-KO cells, using both the AP and NC Connectivity scores between hairpins targeting three antibodies, and were also absent from IgG lysates (Fig. 2a; one-carbon enzymes included in CMap (MTHFD2, Methods). With these criteria, we identified 29 MTHFD1, and SHMT1) and hairpins targeting 3799 ex- MTHFD2-interacting proteins, from here on referred to amined genes were analyzed using the CMap online as the D2PPI gene set (Table 1). The majority of these pro- tools (https://clue.io/cmap). Connectivity scores of more teins have been reported to localize to the nucleus, than 90 or less than −90 were considered as significantly consistent with previous evidence that MTHFD2 is also enriched. Clustering analysis of connectivity scores was anuclearprotein[17]. While two mitochondrial performed using ClustVis [27] with average linkage and RNA-binding proteins were present (LRPPRC; HSPA9), correlation metric. Ribosomal genes were selected for we found no other one-carbon enzymes among the Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 5 of 14

HCT-116 HCT-116 A WT MTHFD2-/- B C 1 2345 15 WT cell lysis MTHFD2-/- 10 70 50 4 2 222 5 MTHFD2 AP NCIgG AP NC 37 25 Unique MTHFD2 peptides MTHFD2 Unique n.d. Immuno- In-gel digestion precipitation AP NC LC-MS

AP 29 NC WT-KO WT-KO

IgG

DFM olecular function

histone binding HSPD1 unfolded protein binding RPS3A HSPA9 structural constituent of ribosome RPS5 HSPA8

structure-specific DNA binding RPS8 HSPB1 RPS13 HNRNPU RNA binding HNRNPM

nucleic acid binding SF3B3 NPM1 0 0 0 RPL18 RPN1 1 20 3 HNRNPK ILF2

E Cellular Component XRCC6 HNRNPR SYNCRIP cytosol RPA1 DDX3X mitochondrial nucleoid HIST1H3A ribosome nucleolus nuclear EZR ENO1 HIST3H3 HIST2H3A H3F3A LRPPRC NUFIP2 focal adhesion nucleus spliceosomal complex Translation Chaperones ribonucle oprote in comple x RNA processing DNA-associated

0 5 5 0 10 1 2 25

Number of Proteins Fig. 2 Defining the MTHFD2 interactome. a Schematic of the strategy used for identifying MTHFD2 interactors. Immunoprecipitation was performed on parental HCT-116 cells (WT) and MTHFD2 knockout cells (D2-KO) using either one of two anti-MTHFD2 antibodies (AP and NC) or an anti-rabbit immunoglobulin (IgG) antibody. The numbers above antibody names indicate the number of replicate experiments performed for a given condition. After in-gel digestion and MS 29 proteins were found to be detected in IP from both anti-MTHFD2 antibodies in parental HCT-116 cells but not in IP performed in MTHFD2 knockout cells or using IgG antibody. b Representative silver-stained SDS-gel containing co-immunoprecipitated samples. Lane 1 contained a molecular weight ladder. Lane 2 and 3 lysates from wildtype HCT-116 cells after immunoprecipitation with anti-MTHFD2 antibody, lanes 4 and 5 lysates from same cells immunoprecipitated using anti-rabbit IgG antibody. c Number of unique peptides matching to MTHFD2 identified in replicate experiments with two anti-MTHFD2 antibodies in WT or knockout D2-KO cells. d, e Selected significantly enriched GO Molecular Function and Cellular Component respectively. f PPI network generated for 29 identified MTHFD2 interactors using experimental evidence in STRING database and medium confidence links (> 0.400). Nodes with no interaction are not shown. Thickness of edges represent strength of evidence supporting interaction. Members of protein functional groups are indicated according to color chart

D2PPI proteins. We did detect MTHFD1 using the AP anti- (GO) analysis of these proteins confirmed body, but this protein was found in D2-KO cells as well, in- a highly significant enrichment of RNA binding proteins − dicating that this interaction is due to unspecific binding. (24/29 proteins; q <10 20)(Fig.2d; Additional file 4)and Hence, MTHFD2 does not appear to participate in the previ- that a large fraction of these proteins were also nu- ously reported nuclear thymidylate synthesis complex [31]. clear (Fig. 2e). To determine if the D2PPI proteins Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 6 of 14

Table 1 MTHFD2 protein interactors identified by CoIP and MS Symbol Name AP NC RNA binding Unfolded DNA Binding protein binding HSPA8 Heat shock cognate 71 kDa protein 3/4 1/2 X X HSPA9 Stress-70 protein, mitochondrial 3/4 1/2 X X HSPB1 Heat shock protein beta-1 3/4 1/2 X HSPD1 60 kDa heat shock protein 3/4 1/2 X X X NPM1 Nucleophosmin 2/4 1/2 X X HIST1H3A Histone H3.1 2/4 1/2 X HIST2H3A Histone H3.2 2/4 1/2 X HIST3H3 Histone H3.1 t 2/4 1/2 X H3F3A Histone H3.3 2/4 1/2 X XRCC6 X-ray repair cross-complementing protein 6 2/4 1/2 X X RPA1 Replication protein A 70 kDa DNA-binding subunit 2/4 1/2 X HNRNPK Heterogeneous nuclear ribonucleoprotein K 2/4 1/2 X X HNRNPM Heterogeneous nuclear ribonucleoprotein M 3/4 1/2 X SYNCRIP Heterogeneous nuclear ribonucleoprotein Q 2/4 1/2 X HNRNPR Heterogeneous nuclear ribonucleoprotein R 2/4 1/2 X HNRNPU Heterogeneous nuclear ribonucleoprotein U 2/4 2/2 X X SF3B3 Splicing factor 3B subunit 3 2/4 1/2 X LRPPRC Leucine-rich PPR motif-containing protein, mitochondrial 3/4 1/2 X X ILF2 Interleukin enhancer-binding factor 2 2/4 1/2 X X NUFIP2 Nuclear fragile X mental retardation-interacting protein 2 2/4 1/2 X DDX3X ATP-dependent RNA helicase DDX3X 2/4 1/2 X X RPL18 60S ribosomal protein L18a 2/4 1/2 X RPS13 40S ribosomal protein S13 2/4 1/2 X RPS3A 40S ribosomal protein S3a 2/4 1/2 X RPS5 40S ribosomal protein S5 2/4 1/2 X RPS8 40S ribosomal protein S8 2/4 1/2 X ENO1 Alpha-enolase 2/4 1/2 X X EZR Ezrin 2/4 1/2 X RPN1 Dolichyl-diphosphooligosaccharide--protein glycosyltransferase subunit 1 2/4 1/2 X Numbers indicate the unique peptides mapped to protein in given CoIP replicate belonged to any known complexes, we extracted a Co-expression analysis of MTHFD2 and interacting proteins protein-protein interaction (PPI) network over these To systematically assess if the D2PPI proteins indeed 29 proteins from the STRING database [21]. We ob- share a common function with MTHFD2, we next asked tained 47 physical interactions, which is two times whether their mRNAs are also frequently co-expressed − more than expected by chance (p <10 7); hence, the with MTHFD2. We have previously shown that frequent D2PPI proteins are likely part of previously known co-expression across diverse experimental conditions is protein complexes, including the small ribosomal sub- highly predictive of closely related biological function unit proteins (four RPS proteins), RNA processing [32, 33]. Using a previously described method [17], we proteins (several hnRNP family members; SF3B3; scored 25,845 genes for co-expression with MTHFD2 in ILF2), protein chaperones (several heat shock pro- 8067 human, rat, and mouse microarray datasets teins; NPM1), and histone/DNA repair proteins (three (Additional file 5). Remarkably, D2PPI gene set was clearly histone proteins; XRCC6; RPN1) (Fig. 2f). Hence, the enriched for coexpression with MTHFD2 (p =0.003, proteins shown here to directly interact with permutation test), with 13 of 29 of genes within the 95th MTHFD2indicatearolefortheproteininmetabol- percentile of coexpression scores (Fig. 3a). An exception ism and translation of RNA. was the histone proteins H3F3A, HIST1H3A, HIST3H3, Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 7 of 14

A B

CDEF

Fig. 3 MTHFD2-interacting proteins are coexpressed with MTHFD2 in response to stimuli. a Enrichment plot showing the ranks of MTHFD2-interacting (D2PPI) proteins ordered by overall coexpression with MTHFD2 across 8067 gene expression data sets (see Methods). b Heat map of coexpression score for the D2PPI and ATF4 gene sets, across the 50 data sets exhibiting with strongest coexpression for each set. Data sets selected for panels c–f are indicated by arrows. c–f Heat maps of expression level for selected data sets. For each gene, ratio over the overall mean is shown, according to scale bars. Only genes with nonzero coexpression scores are shown. MTHFD2 expression level is shown above in linear scale, with baseline equal to zero (arbitrary units) and HIST2H3A that did not exhibit co-expression with example, in one leukemia study, highest expression was MTHFD2; however, these proteins were not well-repre- found in T-ALL and lowest in AML subtypes, suggesting sented on the microarrays used, so this result may reflect specific disease contexts where the D2PPI proteins may low power to detect coexpression. An independent ana- be relevant (Fig. 3c). We also noted a concerted induction lysis of 3356 human gene expression datasets using the of the D2PPI proteins in stimulated T cells (Fig. 3d), in SEEK gene co-expression analysis tool [26]gavesimilar line with previous observations of MTHFD2 induction in results (Additional file 5). To investigate factors that might this setting [4], and also suppression of D2PPI genes in drive the observed coexpression of the D2PPI gene set, we HCT-116 cancer cells in response to treatment with a identified specific gene expression data sets where this set CDK inhibitor (Fig. 3e). These data indicate that was coexpressed (Fig. 3b, Additional file 6). Among the MTHFD2 and the D2PPI genes are frequently top scoring data sets were several tumor gene expression coexpressed and responsive to mitogenic stimuli and studies, where expression varied between subtypes; for anti-proliferative drugs. Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 8 of 14

As previously shown [17], a set of ATF4-responsive in their response to serum depletion and stimulation, mRNAs involved in metabolism and with MTHFD2 being particularly responsive. aminoacyl-tRNA synthesis, including the mitochondrial folate-coupled enzymes SHMT2 and MTHFD1L, was Transcriptional response to MTHFD2 knockdown in also strongly co-expressed with MTHFD2. However, we cancer cells found that these genes co-express with MTHFD2 in dif- To study the effects of loss of MTHFD2, and how these ferent conditions than the MTHFD2-interacting proteins relate to its identified interacting proteins, we analyzed (Fig. 3b, Additional file 6), suggesting that the underlying transcriptomics data from MTHFD2 shRNA knockdown mechanism is different. For example, the mitochondrial cells generated by the Connectivity Map (CMap) project enzymes were induced in contexts of amino acid starva- [34]. Interestingly, the D2PPI gene set was clearly de- − tion, possibly reflecting the involvement of this pathway creased (ES -0.64, p <10 4) in MTHFD2 knockdown in amino acid synthesis, while the D2PPI genes were not cells (Fig. 5a) and was among the most negatively (Fig. 3f). In summary, the D2PPI genes are co-regulated enriched among examined collection of 3581 gene sets with MTHFD2 in specific biological conditions, provid- (Fig. 5b). Hence, loss of MTHFD2 affects expression of ing independent evidence that they share a biological the D2PPI genes, consistent with a shared function. function with MTHFD2, which appears to be distinct The CMap data also allows assessing functional rela- from ATF4-driven induction of MTHFD2. tionship between genes by scoring the similarity of tran- scriptional response of cancer cells to individual gene MTHFD2 is a rapidly regulated protein with a short half-life knockdown. In this regard, knockdown of MTHFD2 was Given the above finding that MTHFD2 and the D2PPI highly similar (score > 90/100) to four out of the 14 D2PPI gene set is acutely regulated by pro- and anti-proliferative genes represented in CMap: HSPA8, HSPA9, HSPD1, and stimuli, we examined in more detail the dynamics of RPS3A. In contrast, with this criterion, other MTHFD2 regulation. Generally, metabolic enzymes are folate-metabolizing enzymes present in the CMap data set long-lived proteins, while proteins involved in RNA me- (MTHFD1, SHMT2, GLDC, AMT, TYMS, MTR, MTRR, tabolism, cell cycle, and signaling have shorter half-lives FTCD, and DHFR) were not similar to MTHFD2, with the [22]. In a reanalysis of global protein half-life data [22, 23], exception of MTHFD1 (score = 91.0, rank 414/3798), pos- we noted that the half-life of the MTHFD2 protein was sibly reflecting the fact that loss of MTHFD1 also gives se- consistently within 17–22 h in HeLa, C2C12, and NIH3T3 vere growth phenotypes [7]. In particular, SHMT2 was cells, which was markedly lower than the other folate not closely related (score = 18.74 of 100, rank 2790/3798), one-carbon enzymes (Fig. 4a) and within the 5–10th per- consistent with the observations that loss of MTHFD2 is centile of all detected enzymes (as defined by Recon2 [24]; often detrimental to cells [4], while SHMT2 is not [2], and 596–605 enzymes were detected across three cell lines) suggesting that at least some effects of MTHFD2 suppres- (Fig. 4b). We confirmed this short half-life of MTHFD2 by sion are unrelated to mitochondrial folate metabolism. treatment with the cyclohexamide inhibitor of protein Finally, we investigated the sets of genes that in CMap synthesis, which resulted in an approximate halving of were highly similar to MTHFD2. We noted that, across all protein levels after 24 h (Fig. 4c). Therefore, MTHFD2 is 3798 genes for which shRNA data was available in the unusually short-lived for a metabolic enzyme in general, CMap data set, MTHFD2 was highly similar (absolute and one-carbon enzymes in particular. score > 90) to 510 genes, compared to only 266 for Considering the short half-life of MTHFD2, we next SHMT2 and 169 for MTHFD1, indicating that loss examined the regulation of the MTHFD2 protein com- MTHFD2 induces a response that more commonly occurs pared to other enzymes in the mitochondrial folate path- with gene knockdown. In addition, the set of 510 genes way. In HCT-116 cells, after 48 h serum starvation, similar to MTHFD2 in this respect were clearly different MTHFD2 and SHMT2 mRNA was significantly de- from those of SHMT2 and MTHFD1 (Fig. 5c), again con- creased, while SHMT1, MTHFD1, and MTHFD1L was sistent with MTHFD2 having a distinct function. In par- unaffected (Fig. 4d). At the protein level, MTHFD2 and ticular, knockdown of MTHFD2, but not SHMT2 or MTHFD1L decreased noticeably at the 24 h time point, MTHFD1, was highly similar to knockdown of a group of while MTHFD1 did not (Fig. 4e), consistent with the ribosomal proteins (Fig. 5d). Therefore, MTHFD2 is shorter half-life of MTHFD2. In serum restimulated unique in that suppression of this protein causes cellular HCT-116 cells, MTHFD2 protein was induced within transcriptomic responses that are highly similar to those 3 h (Fig. 4f), consistent with our previous report [17]. observed following the targeting of ribosomal proteins. Moreover, induction of the protein was observed in both the nuclear and cytoplasmic compartments (Fig. 4h), Comparison of RNAi and CRISPR suppression of MTHFD2 and also on the mRNA level (Fig. 4g). Our data show At first, the observation that MTHFD2 knockdown is therefore that enzymes of the one-carbon pathway differ similar to knockdown of ribosomal proteins, but not the Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 9 of 14

ABC

D E

F G

H

Fig. 4 MTHFD2 is a short-lived and dynamically regulated protein. a Half-life of enzymes of the one-carbon pathway in in HeLa, C2C12, and NIH3T3 cell lines. b Cumulative frequency of all quantified enzymes (596, 576, and 605 respectively) half-life in three cell lines. Dotted lines indicate relative half-life of MTHFD2 in each cell line. c Effect of treatment of HCT-116 cells for 24 h with 25 mg/ml protein translation inhibitor cyclohexamide on MTHFD2 protein levels. d qRT-PCR measurement of effect of 48 h serum starvation on mRNA of folate one-carbon enzymes. e Immunoblots of MTHFD1, MTHFD1L, MTHFD2, and beta-tubulin in HCT-116 grown in normal media or serum starved for 24 or 48 h. f Immunoblot showing time course of MTHFD2 protein response to serum replenishment in HCT-116 cells. SS, serum-starved; rep, serum replenishment (time point indicated). g Quantification of MTHFD2 mRNA in HCT-116 cells in control media, serum starved (SS) for 48 h, and serum starved followed by 24 h serum replenishment (rep). h Levels of MTHFD2 protein in nuclear and cytosolic compartments in HCT-116 cells that were serum starved for 48 h or in replenished media for 24 h after serum starvation. The COX IV and Lamin were used as cytosolic and nuclear markers respectively. Numbers in f and g indicate fold changes relative to control Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 10 of 14

A B 1.0 D2PPI gene set HIST3H3 0.5 RPS5 ES = -0.64 HIST3H3 0

Down Up -0.5 Enrichment score

12,328 genes -1.0 D2PPI

3582 gene sets

CD510 high-scoring genes connectivity 16 ribosomal genes connectivity score score MTHFD2 100 MTHFD2 50 80 60 MTHFD1 SHMT2 0 40 -50 20 SHMT2 MTHFD1 0 -100 RPL22 UBA52 RPL7 RPS15A RPS3 FAU RPS7 RPS27 RPS3A RPS10 RPS9 RPS16 RPS5 RPS6 RPS13 RPS19 A

Fig. 5 Consequences of MTHFD2 knockdown. a Enrichment plot showing ranks of the D2PPI gene set among 12,328 genes responding to treatment with MTHFD2 shRNA in CMap study. Overall enrichment score was − 0.64. Genes with low enrichment scores (histones and RSP5) are indicated. b Distribution of enrichment scores for 3582 gene sets, with arrow indicating position of D2PPI gene set. c Heatmap showing connectivity scores for all 510 genes with significant associations (scores > 90 or < − 90) in CMap project with either of MTHFD2, MTHFD1 or SHMT2. Genes were clustered by average linkage and Pearson correlation distance. d Connectivity scores between MTHFD2, MTHFD1, and SHMT2 with ribosomal genes measured in the CMap project upstream enzyme SHMT2, seemed counter-intuitive. A weak correlations (Fig. 6c). However, in these data sets, possible explanation is that shRNA knockdown cells still the D2PPI proteins frequently showed strong growth contain some residual protein, which may be sufficient phenotypes across all cell lines and were not significantly to retain activity of a metabolic enzyme, but disrupts a correlated with MTHFD2 (data not shown). This is con- more sensitive non-metabolic function such as control sistent with the notion that in CRISPR MTHFD2 knock- of translation. Indeed, there is evidence that a small outs, the metabolic disruption dominates the effects on amount of residual enzymatic activity can be sufficient cell proliferation, while for shRNA knockdown, other to maintain necessary metabolic output, as seen for ex- functions of MTHFD2 may play a role. ample with SHMT2 and formylmethionyl-tRNA pools [35]. To investigate this hypothesis, we compared the ef- Discussion fects of shRNA knockdown to effects of CRISPR knock- A common theme in cancer metabolism is that out, using genome-wide shRNA and CRISPR screening cancer-associated enzymes of interest as potential thera- data from the Achilles project, measuring growth pheno- peutic targets [36–38] are frequently found to have types across 501 and 341 cell lines, respectively. Here, non-metabolic “moonlighting” functions [39, 40]. In this genes with shared functions are expected to have similar study, we have demonstrated that MTHFD2 interacts dependency profiles across cell lines [30]. Overall, 977 with ribosomal and RNA processing proteins (Fig. 2, genes were similar to MTHFD2 in the shRNA screening Table 1), which are also coexpressed with MTHFD2, and data, and 161 with CRISPR screening (Pearson’s correl- give similar transcriptional phenotypes upon shRNA ation, Bonferroni-corrected p < 0.05). Interestingly, genes knockdown. These distinct methods provide independ- involved in folate metabolism (SHMT2, GART, GCSH, ent evidence suggesting that MTHFD2 has a previously FPGS) were similar to MTHFD2 in the CRISPR, but not unrecognized function in RNA metabolism and/or trans- in the shRNA screens (Fig. 6a, b, Additional file 7). lation, although further experiments are required to con- Moreover, in a similar analysis of an independent CRISPR firm these findings, for example using ribosomal dataset comprising 14 AML cell lines, SHMT2, profiling, in-depth analysis of alternative splicing, and SLC25A32, and MTHFD1L were strongly correlated with additional interaction data including reverse Co-IP ex- MTHFD2, while cytosolic one-carbon enzymes showed periments. The MTHFD2-interacting proteins clearly Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 11 of 14

ABCRISPR C 0.2 CRISPR R = 0.27 AML cell lines Gene Pearson R Rank 0 100 GCSH 0.33 4 -0.2 MTHFD1L R = 0.57 GART 0.29 38

SHMT2 -0.4 SHMT2 75 MTHFD2L FPGS 0.28 48 R = 098 -0.6 R = 0.05 SHMT20.27 56 SLC25A32 -0.8 50 R = 0.78 -0.6 -0.4 -0.200.4 0.2 ALDH1L2 SHMT1

Percentile R = -0.26 R = 0.03 MTHFD2 25 MTHFD1 shRNA R = -0.11 shRNA 0 Gene Pearson R Rank 10 R = -0.03 -1.0 -0.5 0 0.5 1.0 GCSH 0.11 5004 5 Correlation GART 0.01 14794 FPGS 0.01 15822 0 SHMT2 -0.03 13686 SHMT2 -5

-10 -4 -204 2 MTHFD2 Fig. 6 Comparison of consequences of CRISPR knockout and RNAi knockdown of MTHFD2. a Pearson correlation and global rank between MTHFD2 and genes involved in folate metabolism in shRNA and CRISPR screens. b Comparison of growth effects caused by depletion of MTHFD2 and SHMT2 using shRNA and CRISPR in Achilles project. Each circle represents a tested cell line. Pearson’s correlation is shown. c Distribution of correlations between growth phenotype effects of all examined genes and MTHFD2 across 14 AML cell lines. Selected genes involved in one-carbon metabolism are indicated by arrows consist of previously known complexes of proteins with the HSPs [49]. Interestingly, some of the observed related, but also diverse, functions. We found multiple hnRNPs are involved in heat shock responses as well members of the heterogeneous ribonuclear protein [50], hinting at a common role. We also found interac- (hnRNP) family of RNA binding proteins, which are of tions with proteins involved in DNA repair and replica- particular interest for several reasons. hnRNPs are tion (XRCC6 and RPN1), as well as histones (Fig. 2; known to interact with nascent RNA and control their Table 1, consistent with observations that the protein stability, localization, splicing, and translation, and sev- localize to regions of newly formed DNA [17]. Whether eral family members have been implicated in regulation these findings represent more than one distinct function of cell proliferation and in the endothelial-mesenchymal of the MTHFD2 protein, or rather are multiple aspects transition (EMT) process [41, 42], which are also af- of a single mechanism, is not yet clear. fected by RNAi against MTHFD2 [42–44]. While the Although Co-IP coupled with MS is a powerful meth- consequences of interaction between MTHFD2 and odology for the unbiased detection of protein interac- hnRNPs are not yet clear, one possibility is that this tions, it is still a technique with false positives (due to interaction may allow nuclear MTHFD2 to exert some off-target binding of antibodies) and false negatives (due control over gene expression; indeed, there are many ex- to failure to capture protein-proteins interactions that amples of hnRNP-interacting proteins influencing gene are of transient nature, or due to other technical rea- regulation via hnRNPs [45–47]. Moreover, hnRNPs may sons). To reduce the impact of false positives, we utilized mediate signals on metabolic status: in particular, two antibodies and a CRISPR knockout control. Due to hnRNP-E1 has been reported to regulate gene transcrip- this stringent approach, it is more probable that we have tion in response to folate starvation [48]. We also ob- missed some real protein interactors rather than identi- served interactions with components of the small fied false positives. Interestingly, we did not detect inter- ribosomal subunit, and with several heat-shock proteins actions between MTHFD2 and other one-carbon (HSPs) that assist with the folding and transportation of metabolism enzymes involved in nuclear dTMP synthe- target proteins. While HSPs could interact with sis (SHMT1, SHMT2, MTHFD1, TYMS or DHFR) [51], MTHFD2 simply to assist its own folding, it is also pos- nor with nuclear lamin to which they are tethered [31]. sible that MTHFD2 or other components of the ob- Such interactions could of course have been missed in served complexes bind and that regulate the functions of our experiments, since we used quite stringent criteria Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 12 of 14

to obtain high-quality interactions and did not perform Additional files cross-linking during CoIP. Alternatively, MTHFD2 may reside in a different compartment. From a biochemical Additional file 1: Table S1. List of primers used for qRT-PCR in this perspective, nuclear MTHFD2 would not be expected to study. (DOCX 12 kb) Additional file 2: Figure S1. Validation of MTHFD2 knockout cell model. contribute to dTMP synthesis, since this process re- (a) Immunoblot of MTHFD2 in WT and D2-KO clones. Tubulin is shown as quires CH2-THF which MTHFD2 cannot generate on its loading control. (b) Immunoprecipitation of MTHFD2 using Anti-MTHFD2 own, as it lacks the formyl-THF synthase domain antibody from D2-KO and WT cells followed by immunoblotting for MTHFD2. (c) Silver-stained SDS-PAGE of CoIP lysates from WT and D2-KO present in MTHFD1. cells. As indicated the CoIP were performed using either of the anti- An open question is what drives the recruitment and MTHFD2 NC or AP antibodies. MTHFD2 bands are indicated by blue association of MTHFD2 with cellular RNA-processing squares. As can be seen MTHFD2 were detected in IP samples from protein complexes. Because several of the D2PPI pro- WT but not D2-KO cells. (PDF 311 kb) Additional file 3: Mapped peptides detected by Mass Spectrometry in teins are RNA-binding, one possibility is that MTHFD2 each of 12 CoIP lysate samples. (XLSX 748 kb) itself is an RNA-binding protein and encounters its part- Additional file 4: Significantly enriched GO Molecular Function and ners while bound to specific . Indeed, a recent Cellular Component gene sets for 29 identified MTHFD2 protein interactors. large-case study reported that RNA binding proteins (XLSX 15 kb) tend to CoIP with a large number of other RNA binding Additional file 5: Co-expression scores between MTHFD2 and the D2PPI genes. The gene co-expression were calculated across 8097 human, rat, and proteins [52]. Additionally, MTHFD2 contains Rossman mouse datasets using in-house tools, or against ~ 3000 human datasets folds, NAD-binding domains that can also bind RNA, as using SEEK. (XLSX 11 kb) in the example of GAPDH [53]. A number of enzymes Additional file 6: List of top data sets exhibiting coexpression between have been shown to “moonlight” by binding RNA, in- MTHFD2 and the D2PPI and ATF4 gene sets, respectively. (XLSX 17 kb) cluding SHMT2 and MTHFD1 [54], and it has been sug- Additional file 7: Correlation scores of MTHFD2 against all other genes gested that this may allow metabolic status to influence in Achilles Project. (XLSX 1006 kb) gene regulation [55]. However, it is also possible that MTHFD2 is not a direct RNA-binding protein, but in- Abbreviations AP: Proteintech 12,270–1-AP anti-MTHFD2 antibody; CoIP: Co-immunoprecipitation; stead contacts one or more proteins around which the D2-KO: MTHFD2 CRISPR knockout cell line; D2PPI: Set of D2 protein-protein RNA-binding complex is organized. Further work is re- interactors, listed in Table 1.; HNRNP: Heterogeneous ribonucleoprotein; quired to demonstrate whether MTHFD2 binds RNA HSP: Heat-shock proteins; IgG: Immunoglubulin G; MS: Mass spectrometry; MTHFD2: Methylenetetrahydrofolate dehydrogenase (NADP+ dependent) species directly or indirectly, and understand how the 2; NC: Genetex N1C3 anti-MTHFD2 antibody; PPI: Protein-protein interaction observed protein-protein interactions are formed and function. Further experiments will also be important for Acknowledgements validating and extending the findings of this study, such We would like to thank Dr. Rabinowitz for providing us with the MTHFD2 CRISPR knockout cells. In-gel digestion, peptide extraction, MS analysis, and as Co-IP and transcriptomic studies of enzymatically in- database searches of protein identification were performed by the Proteomics active MTHFD2 in diverse nutritional and cell condi- Karolinska (PK/KI) core facility at Karolinska Institute, Stockholm. tions. For example it will be of interest to determine whether the interaction partners of MTHFD2 are modi- Funding The project was funded by Cancerfonden (CAN 2016/632) and the Swedish fied in response to availability of one-carbon units, Foundation for Strategic Research (FFL12-0220). growth factors or DNA damage. Availability of data and materials Conclusions The CMap data can be accessed from the project’s online tool (https://clue.io/cmap). Achilles Project data are available from the Broad Institute portal (https:// Our study suggests a role for nuclear MTHFD2 in RNA portals.broadinstitute.org/achilles). Protein half-life data are available from metabolism and translation, besides its established func- original publications [22, 56]. Co-expression with MTHFD2 calculated by tion in mitochondrial folate metabolism. An intriguing SEEK are available online (http://seek.princeton.edu). possibility is that the MTHFD2-interacting proteins could Authors’ contributions provide a mechanism whereby MTHFD2 can affect gene RN and CK conceived the study, designed the experiments, wrote the expression and cell behaviour, perhaps serving to integrate manuscript, and approved for the publication of the final manuscript. information on folate metabolism. While the precise CK performed the experiments. CK and RN performed the computational analysis presented here. mechanism remains to be investigated, such a function would explain the observation that catalytically inactive Ethics approval and consent to participate MTHFD2 protein is sufficient to promote cell prolifera- Not applicable. tion [17]. Future investigation of this non-enzymatic func- tion will be important for pharmaceutically targeting Consent for publication Not applicable. MTHFD2; for example, inhibitors of MTHFD2 dehydro- genase activity may not be effective, and instead drugs that Competing interests disrupt protein-protein interactions might be required. The authors declare that they have no competing interests. Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 13 of 14

Publisher’sNote 19. Schmidt A, Wu H, MacKenzie RE, Chen VJ, Bewly JR, Ray JE, et al. Structures Springer Nature remains neutral with regard to jurisdictional claims in published of three inhibitor complexes provide insight into the reaction mechanism maps and institutional affiliations. of the human methylenetetrahydrofolate dehydrogenase/cyclohydrolase. Biochemistry. 2000;39:6325–35. Author details 20. Fu C, Sikandar A, Donner J, Zaburannyi N, Herrmann J, Reck M, et al. The 1Cardiovascular Medicine Unit, Department of Medicine, Karolinska Institutet, natural product carolacton inhibits folate-dependent C1 metabolism by SE-171 76 Stockholm, Sweden. 2Division of Cardiovascular Medicine, targeting FolD/MTHFD. Nat Commun. 2017;8:1529. Karolinska University Hospital, SE-171 76 Stockholm, Sweden. 3Center for 21. Szklarczyk D, Morris JH, Cook H, Kuhn M, Wyder S, Simonovic M, et al. The Molecular Medicine, Karolinska Institutet, SE-171 76 Stockholm, Sweden. STRING database in 2017: quality-controlled protein–protein association networks, made broadly accessible. Nucleic Acids Res. 2017;45:D362–8. Received: 29 May 2018 Accepted: 8 August 2018 22. Schwanhäusser B, Busse D, Li N, Dittmar G, Schuchhardt J, Wolf J, et al. Global quantification of mammalian gene expression control. Nature. 2011;473:337–42. 23. Cambridge SB, Gnad F, Nguyen C, Bermejo JL, Krüger M, Mann M. Systems- wide proteomic analysis in mammalian cells reveals conserved, functional References protein turnover. J Proteome Res. 2011;10:5275–84. 1. Pavlova NN, Thompson CB. The emerging hallmarks of cancer metabolism. 24. Thiele I, Swainston N, Fleming RMT, Hoppe A, Sahoo S, Aurich MK, et al. Cell Metab. 2016;23:27–47. A community-driven global reconstruction of human metabolism. Nat 2. Jain M, Nilsson R, Sharma S, Madhusudhan N, Kitami T, Souza AL, et al. Biotechnol. 2013;31:419–25. Metabolite profiling identifies a key role for glycine in rapid cancer cell 25. Subramanianan A, Tamayo P, Mootha VK, Mukherjeed S, Ebert BL, Gillette proliferation. Science. 2012;336:1040–4. MA, et al. Gene set enrichment analysis: a knowledge-based approach for 3. Locasale JW. Serine, glycine and one-carbon units: cancer metabolism in full interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. circle. Nat Rev Cancer. 2013;13:572–83. 2005;102:15545–50. 4. Nilsson R, Jain M, Madhusudhan N, Sheppard NG, Strittmatter L, Kampf C, et 26. Zhu Q, Wong AK, Krishnan A, Aure MR, Tadych A, Zhang R, et al. Targeted al. Metabolic enzyme expression highlights a key role for MTHFD2 and the exploration and analysis of large cross-platform human transcriptomic mitochondrial folate pathway in cancer. Nat Commun. 2014;5:3128. compendia. Nat Methods. 2015;12:211–4. 5. Pike ST, Rajendra R, Artzt K, Appling DR. Mitochondrial C1-tetrahydrofolate 27. Metsalu T, Vilo J. ClustVis: a web tool for visualizing clustering of multivariate synthase (MTHFD1L) supports the flow of mitochondrial one-carbon units data using principal component analysis and heatmap. Nucleic Acids Res. into the methyl cycle in embryos. J Biol Chem. 2010;285:4612–20. 2015;43:W566–70. 6. Lewis CA, Parker SJ, Fiske BP, McCloskey D, Gui DY, Green CR, et al. Tracing 28. MeyersRM,BryanJG,McFarlandJM,WeirBA,SizemoreAE,XuH,etal. compartmentalized NADPH metabolism in the cytosol and mitochondria of Computational correction of copy number effect improves specificity of mammalian cells. Mol Cell. 2014;55:253–63. Elsevier Inc CRISPR–Cas9 essentiality screens in cancer cells. Nat Genet. 2017;49: 7. Ducker GSSS, Chen L, Morscher RJJJ, Ghergurovich JMMM, Esposito M, Teng 1779–84. X, et al. Reversal of cytosolic one-carbon flux compensates for loss of the 29. Tsherniak A, Vazquez F, Montgomery PG, Weir BA, Kryukov G, Cowley GS, mitochondrial folate pathway. Cell Metab. Elsevier Inc. 2016;23:1140–53. et al. Defining a cancer dependency map. Cell. 2017;170:564–576.e16. 8. Mejia NR, MacKenzie RE. NAD-dependent methylenetetrahydrofolate 30. Wang T, Yu H, Hughes NW, Liu B, Kendirli A, Klein K, et al. Gene Essentiality dehydrogenase is expressed by immortal cells. J Biol Chem. 1985;260: Profiling Reveals Gene Networks and Synthetic Lethal Interactions with 14616–20. Oncogenic Ras. Cell. 2017;168:890–903.e15. 9. Tedeschi PM, Vazquez A, Kerrigan JE, Bertino JR. Mitochondrial 31. Anderson DD, Woeller CF, Chiang E-P, Shane B, Stover PJ. Serine methylenetetrahydrofolate dehydrogenase (MTHFD2) overexpression is Hydroxymethyltransferase anchors de novo thymidylate synthesis associated with tumor cell proliferation and is a novel target for drug pathway to nuclear Lamina for DNA synthesis. J Biol Chem. 2012;287: development. Mol Cancer Res. 2015;13:1361–6. 7051–62. 10. Pikman Y, Puissant A, Alexe G, Furman A, Chen LM, Frumm SM, et al. Targeting 32. Baughman JM, Nilsson R, Gohil VM, Arlow DH, Gauhar Z, Mootha VK. A MTHFD2 in acute myeloid leukemia. J Exp Med. 2016;213:1285–306. computational screen for regulators of oxidative phosphorylation implicates 11. Di Pietro E, Sirois J, Tremblay ML, MacKenzie RE. Mitochondrial NAD-dependent SLIRP in mitochondrial RNA homeostasis. PLoS Genet. 2009;5:e1000590. methylenetetrahydrofolate dehydrogenase-methenyltetrahydrofolate 33. Nilsson R, Schultz IJ, Pierce EL, Soltis KA, Naranuntarat A, Ward DM, et al. cyclohydrolase is essential for embryonic development. Mol Cell Biol. Discovery of genes essential for heme biosynthesis through large-scale 2002;22:4158–66. gene expression analysis. Cell Metab. 2009;10:119–30. 12. Bolusani S, Young BA, Cole NA, Tibbetts AS, Momb J, Bryant JD, et al. 34. Subramanian A, Narayan R, Corsello SM, Peck DD, Natoli TE, Lu X, et al. Mammalian MTHFD2L encodes a mitochondrial methylenetetrahydrofolate A Next Generation Connectivity Map: L1000 Platform and the First 1,000,000 dehydrogenase isozyme expressed in adult tissues. J Biol Chem. 2011; Profiles. Cell. 2017;171:1437–1452.e17. 286:5166–74. 35. Minton DR, Nam M, McLaughlin DJ, Shin J, Bayraktar EC, Alvarez SW, et al. 13. Liu F, Liu Y, He C, Tao L, He X, Song H, et al. Increased MTHFD2 expression is Serine Catabolism by SHMT2 Is Required for Proper Mitochondrial associated with poor prognosis in breast cancer. Tumor Biol. 2014;35:8685–90. Translation Initiation and Maintenance of Formylmethionyl-tRNAs. Mol Cell. 14. Liu X, Huang Y, Jiang C, Ou H, Guo B, Liao H, et al. 2018;69:610–621.e5. Methylenetetrahydrofolate dehydrogenase 2 overexpression is associated 36. Christofk HR, Vander Heiden MG, Harris MH, Ramanathan A, Gerszten RE, with tumor aggressiveness and poor prognosis in hepatocellular carcinoma. Wei R, et al. The M2 splice isoform of pyruvate kinase is important for Dig Liver Dis. 2016;48:953–60. cancer metabolism and tumour growth. Nature. 2008;452:230–3. 15. Koufaris C, Gallage S, Yang T, Lau C-H, Valbuena GN, Keun HC. Suppression 37. Son J, Lyssiotis CA, Ying H, Wang X, Hua S, Ligorio M, et al. Glutamine of MTHFD2 in MCF-7 breast cancer cells increases glycolysis, dependency supports pancreatic cancer growth through a KRAS-regulated metabolic on exogenous glycine, and sensitivity to folate depletion. J Proteome Res. pathway. Nature. 2013;496:101–5. 2016;15:2618–25. 38. Dang L, White DW, Gross S, Bennett BD, Bittinger MA, Driggers EM, et al. 16. Kim D, Fiske BP, Birsoy K, Freinkman E, Kami K, Possemato RL, et al. SHMT2 Cancer-associated IDH1 mutations produce 2-hydroxyglutarate. Nature. drives glioma cell survival in ischaemia but imposes a dependence on Nature Publishing Group. 2009;462:739–44. glycine clearance. Nature. 2015;520:363–7. 39. Yang W, Lu Z. Pyruvate kinase M2 at a glance. J Cell Sci. 2015;128:1655–60. 17. Gustafsson Sheppard N, Jarl L, Mahadessian D, Strittmatter L, Schmidt A, 40. Yu X, Li S. Non-metabolic functions of glycolytic enzymes in tumorigenesis. Madhusudan N, et al. The folate-coupled enzyme MTHFD2 is a nuclear Oncogene. 2017;36:2629–36. protein and promotes cell proliferation. Sci Rep. 2015;5:15029. 41. Geuens T, Bouhy D, Timmerman V. The hnRNP family: insights into their role 18. Lehtinen L, Ketola K, Mäkelä R, Mpindi J-P, Viitala M, Kallioniemi O, et al. in health and disease. Hum Genet. 2016;135:851–67. High-throughput RNAi screening for novel modulators of vimentin 42. Xu Y, Gao XD, Lee J-H, Huang H, Tan H, Ahn J, et al. Cell type-restricted expression identifies MTHFD2 as a regulator of breast cancer cell migration activity of hnRNPM promotes breast cancer metastasis via regulating and invasion. Oncotarget. 2013;4:48–63. alternative splicing. Genes Dev. 2014;28:1191–203. Koufaris and Nilsson Cancer & Metabolism (2018) 6:12 Page 14 of 14

43. Zhang Z, Zhou C, Chang Y, Zhang Z, Hu Y, Zhang F, et al. Long non-coding RNA CASC11 interacts with hnRNP-K and activates the WNT/ß-catenin pathway to promote growth and metastasis in colorectal cancer. Cancer Lett. 2016;376:62–73. 44. Lai C-H, Huang Y-C, Lee J-C, Tseng JT-C, Chang K-C, Chen Y-J, et al. Translational upregulation of aurora-a by hnRNP Q1 contributes to cell proliferation and tumorigenesis in colorectal cancer. Cell Death Dis. 2017;8:e2555. 45. Ji Y, Tulin AV. Poly (ADP-ribose) controls DE-cadherin-dependent stem cell maintenance and oocyte localization. Nat Commun. 2012;3:760. 46. Lee P-T, Liao P-C, Chang W-C, Tseng JT. Epidermal growth factor increases the interaction between nucleolin and heterogeneous nuclear ribonucleoprotein K/poly (C) binding protein 1 complex to regulate the gastrin mRNA turnover. Mol Biol Cell. 2007;18:5004–13. 47. Leal G, Comprido D, de Luca P, Morais E, Rodrigues L, Mele M, et al. The RNA-Binding Protein hnRNP K Mediates the Effect of BDNF on Dendritic mRNA Metabolism and Regulates Synaptic NMDA Receptors in Hippocampal Neurons. eNeuro. 2017;4:ENEURO.0268-17.2017. 48. Tang Y-S, Khan RA, Xiao S, Hansen DK, Stabler SP, Kusumanchi P, et al. Evidence favoring a positive feedback loop for physiologic auto upregulation of hnRNP-E1 during prolonged folate deficiency in human placental cells. J Nutr. 2017;147:482–98. 49. Calderwood SK, Gong J. Heat shock proteins promote cancer: It’s a protection racket. Trends Biochem Sci. 2016;41:311–23. 50. Weighardt F, Cobianchi F, Cartegni L, Chiodi I, Villa A, Riva S, et al. A novel hnRNP protein (HAP/SAF-B) enters a subset of hnRNP complexes and relocates in nuclear granules in response to heat shock. J Cell Sci. 1999; 112(Pt 10):1465–76. 51. Field MS, Kamynina E, Agunloye OC, Liebenthal RP, Lamarre SG, Brosnan ME, et al. Nuclear enrichment of folate cofactors and methylenetetrahydrofolate dehydrogenase 1 (MTHFD1) protect de novo thymidylate biosynthesis during folate deficiency. J Biol Chem. 2014;289:29642–50. 52. Brannan KW, Jin W, Huelga SC, Banks CAS, Gilmore JM, Florens L, et al. SONAR discovers RNA-binding proteins from analysis of large-scale protein- protein Interactomes. Mol Cell. 2016;64:282–93. 53. Nagy E, Rigby WF. Glyceraldehyde-3-phosphate dehydrogenase selectively binds AU-rich RNA in the NAD(+)-binding region (Rossmann fold). J Biol Chem. 1995;270:2755–63. 54. Castello A, Fischer B, Eichelbaum K, Horos R, Beckmann BM, Strein C, et al. Insights into RNA biology from an atlas of mammalian mRNA-binding proteins. Cell. 2012;149:1393–406. 55. Hentze MW, Preiss T. The REM phase of gene regulation. Trends Biochem Sci. 2010;35:423–6. 56. Boisvert F-M, Ahmad Y, Gierlinski M, Charrière F, Lamont D, Scott M, et al. A quantitative spatial proteomics analysis of proteome turnover in human cells. Mol Cell Proteomics. 2012;11:M111.011429.